Image analysis – Applications – Personnel identification
Reexamination Certificate
2000-06-08
2004-11-16
Dastouri, Mehrdad (Department: 2623)
Image analysis
Applications
Personnel identification
C382S203000, C382S218000
Reexamination Certificate
active
06819782
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to devices and methods for recognizing hand shape and position, and recording media each having a program for carrying out the methods recorded thereon, and more specifically to a device and a method for recognizing hand shape and position, without the help of an exemplary cable-connected data glove, in an applicable manner to man-machine interfaces and sign language recognition devices, for example, and to a recording medium having a program for carrying out the method recorded thereon.
2. Description of the Background Art
For a new human interface technique, currently, research and development of a device which recognizes human hand shape and grasps information conveyed thereby is actively conducted. Also, research for recognizing hand shape and position observed in sign language is also active to support communications between the hearing impaired and the able-bodied.
A general method for capturing human hand shape uses a sensor such as data glove to measure hand position and finger joint angles, and an exemplary well-known method is found in the document published by The Institute of Electrical Engineers of Japan, Instrumentation and Measurement (pp. 49 to 56, 1994) (hereinafter, referred to as first document). In the first document, the glove is provided with optical fibers along every finger, and finger joint angles are estimated by a change in light intensity.
A method for recognizing hand shape without the glove-type sensor as in the first document but with a camera is found in the document titled “Gesture Recognition Using Colored Gloves” by Watanabe, et al., (Publication of The Electronic Information Communications Society, Vol. J80-D-
2
, No. 10, pp. 2713 to 2722) (hereinafter, referred to as second document). In the second document, images are captured through a multicolored glove (marker) for hand shape recognition.
An exemplary method for recognizing hand shape and position without such marker but with only a camera is disclosed in the Japanese Patent Laying-Open No. 8-263629 (96-263629) titled “Object Shape/Position Detector” (hereinafter, referred to as third document). In the third document, hand shape recognition and hand position estimation are conducted through images captured by a camera placed in front of a hand. Herein, the method uses at least three cameras to photograph the hand, and the hand is taken in as a plane so as to determine to which camera the hand is facing.
Another method for recognizing hand shape from images captured by a front-facing camera is found in the document titled “Real-Time Vision-Based Hand Gesture Estimation For Human-Computing Interfaces” by Ishibuchi, et al., (Publication of The Electronic Information Communications Society, Vol. J79-D-2, No. 7, paragraphs 1218 to 1229) (hereinafter, referred to as fourth document). In the fourth document, from hand images captured by a plurality of cameras, a direction from wrist to middle finger (hereinafter, referred to as palm principal axis) is determined. And the position of each fingertip is also determined to count the number of extended fingers.
In recent years, to recognize object position and type of face or car, for example, an image recognition method, which is the combination of a dummy image method and an eigenspace method, has been in the spotlight. The dummy image method uses only previously-captured 2D dummy images of a 3D object to recognize the position and type thereof. The eigenspace method is the one conventionally applied, and uses an eigenspace structured by eigenvectors in a covariance matrix (or auto correlation matrix) obtained through an operation performed on a matrix being image data. In the eigenspace method, it is well-known to apply principal component analysis or KL expansion to images.
A technique for applying the principal component analysis to images is briefly described next below.
The principal component analysis is a statistical technique utilizing an eigenspace. This is popular as a technique in multivariate analysis, and is so carried out that featured points on a multidimensional space are represented on a space where the number of dimensions is reduced. This is done to make the featured points easier to see and handle. Fundamentally, featured points on a multidimensional space are linearly projected onto a less-dimensional orthogonal subspace where a distribution level is high.
In a case where the principal component analysis technique is applied to images, first, an image unit including p-piece images is expressed by
{U
1
, U
2
, U
3
, . . . , U
p
},
where U denotes a column vector obtained by subjecting images of n×m pixels to raster scanning.
Second, a component of average image c obtained from a plurality of images is deducted from the respective column vectors in the image unit. Assuming that an nm×p matrix structured by such column vectors is A, the matrix A is expressed by
A=[U
1
−c, U
2
−c, . . . , U
p
−c],
and accordingly a covariance matrix Q is calculated by the following equation (1). Note that, a matrix A
T
indicates a matrix transposed from the matrix A.
Q=AA
T
(1)
Thereafter, a characteristic equation (2) is solved by using the covariance matrix Q.
&lgr;
i
=Qe
i
(2)
Herein, assuming that the number of dimensions of a to-be-structured subspace is k, the subspace can be structured by using eigenvectors which correspond to k-piece large eigenvalues
e
1
, e
2
, . . . , e
k
(&lgr;
1
≧&lgr;
2
≧ . . . ≧&lgr;
k
≧ . . . ≧&lgr;
p
)
as basis.
In this manner, according to the following equation (3), by linearly projecting a certain image x onto the subspace represented by the eigenvectors, the image in the n×m dimension can be represented by a kth dimension featured vector y in a less-dimensional space.
y=[e
1
, e
2
, . . . , e
k
]
T
x
(3)
An exemplary method for detecting and recognizing any multifeatured entity such as human face under principal component analysis or KL expansion is found in the Japanese Patent Laying-Open No. 8-339445 (96-339445) titled “Detection, Recognition and Coding of Complex Objects Using Probabilistic Eigenspace Analysis” (hereinafter, referred to as fifth document). The feature of the fifth document lies in a respect that the conventionally-known principal component analysis and KL expansion are applied to a multifeatured entity such as face. The fifth document exemplarily applies such techniques to recognize hand shape, and the method in the fifth document is described next below.
First, a plurality of hand images captured through hand movement or gesture are photographed with a black background. Second, the two-dimensional contour of the hand is extracted by using Canny's edge operator. Thereafter, the obtained edge images are subjected to the KL expansion to calculate a subspace. If an edge map in binary is used herein, however, the images may show little correlation with one another, and thus the number of dimensions k of the subspace needs to be increased to a considerable extent. By taking this into consideration, the example described in the fifth document proposes to calculate the subspace after blurring the edge images, on the edge map in binary, through distribution processing. In this manner, the number of dimensions of the subspace can be suppressed. Further, in the fifth document, the images are entirely searched on a predetermined size basis so as to find the hand location from an input image, and then recognition is carried out.
However, for hand shape recognition, wearing such data glove as in the first document may restrict hand movement due to codes connected thereto, and a user may feel uncomfortable about wearing the tight glove.
In a case where hand shape recognition is conducted by using a camera presumably together with a marker such as glove, as in the second document, the hand shape recognition cannot be achieved without the glove, and the problem of unc
Igi Seiji
Imagawa Kazuyuki
Lu Shan
Matsuo Hideaki
Dastouri Mehrdad
Matsushita Electric - Industrial Co., Ltd.
Wenderoth , Lind & Ponack, L.L.P.
LandOfFree
Device and method for recognizing hand shape and position,... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Device and method for recognizing hand shape and position,..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Device and method for recognizing hand shape and position,... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3309189